CN113159414A - Traffic speed prediction method based on timing diagram neural network - Google Patents
Traffic speed prediction method based on timing diagram neural network Download PDFInfo
- Publication number
- CN113159414A CN113159414A CN202110417357.XA CN202110417357A CN113159414A CN 113159414 A CN113159414 A CN 113159414A CN 202110417357 A CN202110417357 A CN 202110417357A CN 113159414 A CN113159414 A CN 113159414A
- Authority
- CN
- China
- Prior art keywords
- layer
- time sequence
- time
- attention
- traffic
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 35
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 21
- 238000010586 diagram Methods 0.000 title claims abstract description 13
- 230000002457 bidirectional effect Effects 0.000 claims abstract description 44
- 230000009466 transformation Effects 0.000 claims abstract description 15
- 238000012549 training Methods 0.000 claims description 13
- 230000004927 fusion Effects 0.000 claims description 12
- 230000006870 function Effects 0.000 claims description 6
- 238000010606 normalization Methods 0.000 claims description 6
- 230000008569 process Effects 0.000 claims description 6
- 230000008859 change Effects 0.000 claims description 5
- 239000013598 vector Substances 0.000 claims description 4
- 230000004913 activation Effects 0.000 claims description 3
- 230000009916 joint effect Effects 0.000 claims description 3
- 230000003068 static effect Effects 0.000 abstract description 3
- 230000009977 dual effect Effects 0.000 description 3
- 238000011144 upstream manufacturing Methods 0.000 description 3
- 206010039203 Road traffic accident Diseases 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 238000005295 random walk Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000002123 temporal effect Effects 0.000 description 1
- 238000002759 z-score normalization Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/10—Services
- G06Q50/26—Government or public services
-
- G—PHYSICS
- G08—SIGNALLING
- G08G—TRAFFIC CONTROL SYSTEMS
- G08G1/00—Traffic control systems for road vehicles
- G08G1/01—Detecting movement of traffic to be counted or controlled
- G08G1/0104—Measuring and analyzing of parameters relative to traffic conditions
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Strategic Management (AREA)
- Human Resources & Organizations (AREA)
- Economics (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Development Economics (AREA)
- Health & Medical Sciences (AREA)
- Marketing (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Primary Health Care (AREA)
- Software Systems (AREA)
- Data Mining & Analysis (AREA)
- Computational Linguistics (AREA)
- Evolutionary Computation (AREA)
- Educational Administration (AREA)
- Biophysics (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Molecular Biology (AREA)
- Chemical & Material Sciences (AREA)
- Analytical Chemistry (AREA)
- Game Theory and Decision Science (AREA)
- Entrepreneurship & Innovation (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Traffic Control Systems (AREA)
- Compression Or Coding Systems Of Tv Signals (AREA)
Abstract
The invention discloses a traffic speed prediction method based on a timing diagram neural network, which comprises the following steps of: s1, collecting traffic speed sensor network observation data, and constructing a traffic map and a speed observation sequence; s2, the encoding end carries out feature transformation on the original node features; s3, fusing node space characteristics; s4, modeling dynamic time sequence characteristics in the network based on the bidirectional time sequence space coding layer; s5, the decoding end carries out feature transformation on the original node features; s6, learning coding time sequence characteristics of splicing characteristics based on a bidirectional GRU layer; and S7, calculating the attention between the current state and a plurality of observation states at the encoding end based on the time sequence multi-head attention layer, and predicting. The method solves the problem of modeling of time sequence characteristics and space characteristics in the time sequence traffic network of the static topology, improves the capturing capability of a traffic speed prediction model on the space characteristics and the time sequence dependence characteristics based on the time sequence neural network, and has better usability.
Description
Technical Field
The invention belongs to the technical field of traffic network information, and particularly relates to a traffic speed prediction method based on a timing diagram neural network.
Background
In many cities, especially in big cities of developing countries, the existing transportation system is increasingly burdened with the rapidly expanding demands of private automobiles and trips, and the living standard and urban development of people are seriously affected by traffic jam, long-time commuting and traffic accidents, so that the construction of the intelligent transportation system is more and more emphasized. Traffic speed prediction is an important link in intelligent traffic, and the prediction on the traffic network has the following challenges:
(1) complex spatial dependencies: the roads at different positions in the traffic network have different transportation pressures, for example, on a main road, a downstream road is congested, the running speeds and the flow rates of a middle road and an upstream road are similar, after the downstream road is unblocked, along with the development of time, the condition of the downstream road can be slowly close to that of the upstream road with a higher speed, and finally, the upstream road, the middle road and the downstream road are all restored to a normal state, and the road characteristics of the three roads are also most similar. In addition, in a city traffic network in a dense residential area or an office residential area, traffic congestion and other problems are more likely to occur than in other places.
(2) Dynamically changing time dependence: at the same position, the speed value monitored by the speed sensor can change in time to present nonlinear change, such as a sudden traffic accident at a certain position, which can cause the traffic flow and the traffic speed at the position to rapidly drop. In addition, research shows that for a target position at a certain moment needing to be predicted, the traffic state in a long time in history can show stronger correlation in time sequence with the target position at the current moment than the traffic state in a short time.
Disclosure of Invention
The invention mainly aims to overcome the defects and shortcomings of the prior art, and provides a traffic speed prediction method based on a time sequence diagram neural network, which overcomes the problems of high-order spatial feature modeling of nodes and dual influence mechanics learning of the node features and a topological spatial structure in a short time.
In order to achieve the purpose, the invention adopts the following technical scheme:
a traffic speed prediction method based on a time sequence diagram neural network comprises the following steps:
s1, collecting traffic speed sensor network observation data, and constructing a traffic map and a speed observation sequence;
s2, the encoding end carries out feature transformation on the original node features, and the original node features are transformed into a high-dimensional space by using a full connection layer;
s3, realizing node space feature fusion based on a network representation learning algorithm and a graph convolution neural network;
s4, modeling dynamic time sequence characteristics in the network based on the bidirectional time sequence space coding layer;
s5, the decoding end carries out feature transformation on the original node features;
s6, generating prediction results of a plurality of time points based on the time sequence characteristics of the bidirectional GRU layer learning splicing characteristics;
and S7, calculating the attention between the current state and a plurality of observation states at the encoding end based on the time sequence multi-head attention layer, and predicting.
Further, modeling the space-time characteristics in the traffic speed prediction by adopting an encoding-decoding framework;
the coding end input layer carries out feature transformation on the original node features; in a spatial feature fusion layer, modeling is carried out aiming at spatial feature information of each node in a network, the spatial features of the nodes are captured by utilizing a network representation learning algorithm and a graph convolution neural network, and then the spatial features are fused with the original input features of the nodes; inputting the fused features into a bidirectional time sequence space coding layer, capturing time sequence feature information in a network, and obtaining an output state of a coding end and a hidden state input to the next moment;
the decoding end input layer converts the original characteristics of the nodes in the training stage into a high-dimensional space through a full connection layer; inputting the characteristics, the attention result and the hidden state of the previous moment obtained by the input layer into the bidirectional GRU coding layer, coding the time sequence characteristics, and generating prediction results of a plurality of time points; and calculating the attention value between the intermediate prediction result at the current moment and a plurality of historical observation values at the encoding end through a time sequence multi-head attention layer.
Further, the step S1 specifically includes:
s11, calculating Euclidean distances between every two nodes, calculating a normalized distance through a Gaussian kernel function with a threshold value, and constructing an edge at two nodes corresponding to the normalized distance larger than the threshold value;
s12, repeating the step S11 to obtain a directed traffic map with the right;
and S13, generating traffic speed observation and prediction sequence data with the lengths of T and T' according to the sliding window of the corresponding sliding step length and the sliding time interval for the observation data, and then normalizing the original speed characteristics.
Further, the step S3 specifically includes:
s31, pre-training the traffic map based on the map representation learning algorithm Node2vec to obtain the network space structure embedded representation feature of each Node;
s32, further extracting high-order neighborhood space characteristics of nodes on the traffic map based on the GCN, and performing nonlinear transformation on the embedded expression of the output nodes by using a tanh activation function;
s33, repeating the step S32 for multiple times, and constructing multilayer graph convolution operation;
and S34, carrying out addition fusion with the node characteristics transformed by the full connection layer to obtain the node characteristics with high-order neighborhood space structure information.
Further, the step S4 specifically includes:
s41, learning the time dependence relationship between the current time and the previous time through the bidirectional GRU to obtain the output state and the hidden state of the current time;
s42, before the hidden state obtained in the step S41 is really input to the bidirectional GRU module at the next moment, a graph attention module with residual error connection is used for capturing dynamic features generated by joint action of node features and spatial feature information thereof in the time sequence change process;
s43, carrying out layer normalization on the output of the residual image attention module;
s44, repeating the steps S41 to S42, and stacking two layers of bidirectional GRU networks;
and S45, obtaining the output state of the encoding end for the time sequence multi-head attention layer of the decoding end to calculate attention.
Further, the step S5 specifically includes:
s51, the input layer of the decoding end only plays a role in the model training stage, and the hidden state or the real input value at the previous moment is selected as the input of the bidirectional GRU layer at the current moment with a certain probability in the model training stage;
s52, and in the model prediction phase, the input of each time of the bidirectional GRU layer directly uses the hidden state of the previous time, wherein the first decoded time takes the all-zero vector as the input characteristic of the node.
Further, the step S6 specifically includes:
s61, splicing the features and the attention results obtained by the input layer of the decoding end and the hidden state at the previous moment, inputting the spliced features and the attention results into the bidirectional GRU layer as a whole, and calculating to obtain the hidden state at the next moment and the output at the current moment;
s62, carrying out layer normalization on the output state of the bidirectional GRU layer;
and S63, repeating the steps S61 to S62, and stacking two layers of bidirectional GRU coding networks.
Further, the step S7 specifically includes:
s71, calculating attention values between the traffic state at the current moment and all observation states of the encoding end;
and S72, splicing the attention result, the input layer node characteristics and the output state of the bidirectional GRU layer, outputting the prediction value matrixes of all nodes at the current moment through a full-connection layer, and summarizing to obtain a prediction result sequence with the length of T'.
Compared with the prior art, the invention has the following advantages and beneficial effects:
1. aiming at the problem of high-order spatial feature modeling of a Node in a time sequence traffic network with static topology, the method combines a Node2vec and a graph convolution module of a GCN, utilizes a Node2vec graph representation learning method to pre-train to obtain network structure embedded representation features, and then utilizes the GCN to extract higher-order neighborhood structure information, so that the learning capability of the model on the high-order spatial feature information of the traffic Node is improved.
2. Aiming at the problem of mechanical learning of dual influences of node characteristics and a topological space structure in a time sequence traffic network of static topology, the method adopts a bidirectional time sequence space coding layer with residual map attention, and performs graph convolution operation on an embedded representation of an output node at the current moment by combining a bidirectional GRU and a residual map attention module, so that the node attention under the dual actions of the node characteristics and the topological space structure in the short time is captured, and the capability of a model for capturing dynamic characteristics generated by the spatial characteristics of nodes between adjacent moments is improved.
3. In order to improve the influence of the traffic network state of the key moment on the predicted moment and relieve the error transmission of the model, the method utilizes time sequence multi-head attention to calculate the attention values of each predicted moment of the decoding end and all observation moments of the encoding end, so that the model can pay more attention to the traffic state which has larger influence on the predicted result in the time sequence context, and the learning capacity of the decoding end on the long-distance time dependence characteristic is improved.
Drawings
FIG. 1 is a flow chart of the method of the present invention;
FIG. 2 is a diagram of a model encoding end framework of an embodiment of the present invention;
FIG. 3 is a schematic view of a spatial feature fusion layer of an embodiment of the present invention;
FIG. 4 is a schematic diagram of a bi-directional temporal spatial coding layer according to an embodiment of the present invention;
FIG. 5 is a block diagram of a model decoding end of an embodiment of the present invention.
Detailed Description
The present invention will be described in further detail with reference to examples and drawings, but the present invention is not limited thereto.
Examples
In the embodiment, the method adopts a coding and decoding frame to model the space-time characteristics in the traffic speed prediction, and a coding end input layer carries out characteristic transformation on the original node characteristics; in a spatial feature fusion layer, modeling is mainly performed on spatial feature information of each node in a network, a network representation learning algorithm and a graph convolution neural network are used for capturing the spatial features of the nodes, and then the spatial features are fused with the original input features of the nodes; inputting the fused features into a bidirectional time sequence space coding layer, capturing time sequence feature information in a network, and obtaining an output state of a coding end and a hidden state input to the next moment; the decoding end input layer converts the original characteristics of the nodes in the training stage into a high-dimensional space through a full connection layer; inputting the characteristics, the attention result and the hidden state of the previous moment obtained by the input layer into the bidirectional GRU coding layer, coding the time sequence characteristics, and generating prediction results of a plurality of time points; and calculating the attention value between the intermediate prediction result at the current moment and a plurality of historical observation values at the encoding end through a time sequence multi-head attention layer.
As shown in fig. 1, the method specifically comprises the following steps:
s1, collecting the traffic speed sensor network observation data, and constructing a traffic map and a speed observation sequence, wherein the method specifically comprises the following steps:
longitude and latitude data and speed observation data of a traffic speed sensor network are collected, a traffic map and a speed observation sequence are constructed, Euclidean distances between all nodes are calculated, then a normalized distance is obtained through a Gaussian kernel function with a threshold k, two nodes corresponding to the normalized distance with the threshold k being larger than 0.1 construct an edge, and finally a weighted directed traffic map is obtained; for observation data, traffic speed observation and prediction sequence data of 12 lengths are generated in a sliding window with a sliding step size of 1 and a sliding time interval of 5 minutes, and then raw speed features are normalized by using a z-score normalization method.
S2, the encoding end carries out feature transformation on the original node features, and the original node features are transformed into a high-dimensional space by using a full connection layer; in the present embodiment, the dimension after feature transformation is set to 64; as shown in fig. 2, it is a diagram of the encoding end frame.
S3, realizing node spatial feature fusion based on a network representation learning algorithm and a graph convolution neural network, wherein the node spatial feature fusion is a spatial feature fusion layer as shown in FIG. 3;
in this embodiment, step S3 specifically includes:
s31, pre-training the traffic map based on a graph representation learning algorithm Node2vec to obtain the network space structure embedded representation characteristics of each Node, setting the parameter p to be 2, setting the parameter q to be 1, setting the length of a random walk sequence to be 80, setting the number of the random walk sequences to be 100, outputting a Node embedded dimension to be 64, setting the size of a context window to be 10, and obtaining Node embedded representation after 1000 times of iterative training;
s32, further extracting high-order neighborhood space characteristics of nodes on the traffic map based on the GCN, carrying out nonlinear transformation on embedded expression of output nodes by using a tanh activation function during output, and setting hidden layer dimensions to be 64;
s33, repeating the step S32, and constructing a two-layer graph convolution network;
and S34, carrying out addition fusion with the node characteristics transformed by the full connection layer to obtain the node characteristics with high-order neighborhood space structure information.
S4, modeling the dynamic time sequence characteristics in the network based on the bidirectional time sequence space coding layer, as shown in FIG. 4, the dynamic time sequence characteristics are bidirectional time sequence space coding layers;
in this embodiment, step S4 specifically includes:
s41, learning the time dependence relationship between the current time and the previous time through the bidirectional GRU to obtain the output state and the hidden state of the current time, wherein the number of the hidden units of the GRU is 64;
s42, before the hidden state is really input to the bidirectional GRU module at the next moment, a graph attention module with residual connection is used for capturing dynamic features generated by joint action of node features and spatial feature information of the node features in the time sequence change process, wherein the node feature vectors and the attention value are calculated by using Dropout operation to randomly discard part of network layer output results, the Dropout probability is set to be 0.5, and the number of heads of the graph attention is 8;
s43, carrying out layer normalization on the output of the residual image attention module;
s44, repeating the steps S41 to S42, and stacking two layers of bidirectional GRU networks.
S5, the decoding end carries out feature transformation on the original node features, and is a decoding end frame diagram as shown in FIG. 5;
in this embodiment, step S5 specifically includes:
s51, the input layer of the decoding end only plays a role in the model training stage, and the hidden state or the real input value at the previous moment is selected as the input of the bidirectional GRU layer at the current moment with the probability of 0.5 in the model training stage;
s52, and in the model prediction phase, the input of each time of the bidirectional GRU layer directly uses the hidden state of the previous time, wherein the first decoded time takes the all-zero vector as the input characteristic of the node.
S6, as shown in fig. 5, generating prediction results at a plurality of time points based on the coding time sequence feature of the bidirectional GRU layer learning concatenation feature, in this embodiment, the method specifically includes:
s61, splicing the features and the attention results obtained by the input layer of the decoding end and the hidden state at the previous moment, inputting the spliced features and the attention results into the bidirectional GRU layer as a whole, and calculating to obtain the hidden state at the next moment and the output at the current moment, wherein the number of hidden units of the GRU is 64;
s62, carrying out layer normalization on the output state of the bidirectional GRU layer;
and S63, repeating the steps S61 to S62, and stacking two layers of bidirectional GRU coding networks.
S7, as shown in fig. 5, calculating attentiveness between the current state and a plurality of observation states at the encoding end based on the time-series multi-head attentiveness layer, and predicting, in this embodiment, the method specifically includes:
s71, calculating attention values between the traffic state at the current moment and all observation states at the encoding end, wherein the number of the multiple heads of attention is 4;
and S72, splicing the attention result, the input layer node characteristics and the output state of the bidirectional GRU layer, outputting the prediction value matrixes of all nodes at the current moment through a full-connection layer, and summarizing to obtain a prediction result sequence with the length of 12.
It should also be noted that in this specification, terms such as "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other identical elements in a process, method, article, or apparatus that comprises the element.
The previous description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the invention. Thus, the present invention is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
Claims (8)
1. A traffic speed prediction method based on a time sequence diagram neural network is characterized by comprising the following steps:
s1, collecting traffic speed sensor network observation data, and constructing a traffic map and a speed observation sequence;
s2, the encoding end carries out feature transformation on the original node features, and the original node features are transformed into a high-dimensional space by using a full connection layer;
s3, realizing node space feature fusion based on a network representation learning algorithm and a graph convolution neural network;
s4, modeling dynamic time sequence characteristics in the network based on the bidirectional time sequence space coding layer;
s5, the decoding end carries out feature transformation on the original node features;
s6, generating prediction results of a plurality of time points based on the time sequence characteristics of the bidirectional GRU layer learning splicing characteristics;
and S7, calculating the attention between the current state and a plurality of observation states at the encoding end based on the time sequence multi-head attention layer, and predicting.
2. The traffic speed prediction method based on the neural network of the time sequence diagram according to claim 1, characterized in that a coding-decoding framework is adopted to model the space-time characteristics in the traffic speed prediction;
the coding end input layer carries out feature transformation on the original node features; in a spatial feature fusion layer, modeling is carried out aiming at spatial feature information of each node in a network, the spatial features of the nodes are captured by utilizing a network representation learning algorithm and a graph convolution neural network, and then the spatial features are fused with the original input features of the nodes; inputting the fused features into a bidirectional time sequence space coding layer, capturing time sequence feature information in a network, and obtaining an output state of a coding end and a hidden state input to the next moment;
the decoding end input layer converts the original characteristics of the nodes in the training stage into a high-dimensional space through a full connection layer; inputting the characteristics, the attention result and the hidden state of the previous moment obtained by the input layer into the bidirectional GRU coding layer, coding the time sequence characteristics, and generating prediction results of a plurality of time points; and calculating the attention value between the intermediate prediction result at the current moment and a plurality of historical observation values at the encoding end through a time sequence multi-head attention layer.
3. The method for predicting traffic speed based on the neural network of the time chart according to claim 1, wherein the step S1 specifically comprises:
s11, calculating Euclidean distances between every two nodes, calculating a normalized distance through a Gaussian kernel function with a threshold value, and constructing an edge at two nodes corresponding to the normalized distance larger than the threshold value;
s12, repeating the step S11 to obtain a directed traffic map with the right;
and S13, generating traffic speed observation and prediction sequence data with the lengths of T and T' according to the sliding window of the corresponding sliding step length and the sliding time interval for the observation data, and then normalizing the original speed characteristics.
4. The method for predicting traffic speed based on the neural network of the time chart according to claim 1, wherein the step S3 specifically comprises:
s31, pre-training the traffic map based on the map representation learning algorithm Node2vec to obtain the network space structure embedded representation feature of each Node;
s32, further extracting high-order neighborhood space characteristics of nodes on the traffic map based on the GCN, and performing nonlinear transformation on the embedded expression of the output nodes by using a tanh activation function;
s33, repeating the step S32 for multiple times, and constructing multilayer graph convolution operation;
and S34, carrying out addition fusion with the node characteristics transformed by the full connection layer to obtain the node characteristics with high-order neighborhood space structure information.
5. The method for predicting traffic speed based on the neural network of the time chart according to claim 1, wherein the step S4 specifically comprises:
s41, learning the time dependence relationship between the current time and the previous time through the bidirectional GRU to obtain the output state and the hidden state of the current time;
s42, before the hidden state obtained in the step S41 is really input to the bidirectional GRU module at the next moment, a graph attention module with residual error connection is used for capturing dynamic features generated by joint action of node features and spatial feature information thereof in the time sequence change process;
s43, carrying out layer normalization on the output of the residual image attention module;
s44, repeating the steps S41 to S42, and stacking two layers of bidirectional GRU networks;
and S45, obtaining the output state of the encoding end for the time sequence multi-head attention layer of the decoding end to calculate attention.
6. The method for predicting traffic speed based on the neural network of the time chart according to claim 1, wherein the step S5 specifically comprises:
s51, the input layer of the decoding end only plays a role in the model training stage, and the hidden state or the real input value at the previous moment is selected as the input of the bidirectional GRU layer at the current moment with a certain probability in the model training stage;
s52, and in the model prediction phase, the input of each time of the bidirectional GRU layer directly uses the hidden state of the previous time, wherein the first decoded time takes the all-zero vector as the input characteristic of the node.
7. The method for predicting traffic speed based on the neural network of the time chart according to claim 1, wherein the step S6 specifically comprises:
s61, splicing the features and the attention results obtained by the input layer of the decoding end and the hidden state at the previous moment, inputting the spliced features and the attention results into the bidirectional GRU layer as a whole, and calculating to obtain the hidden state at the next moment and the output at the current moment;
s62, carrying out layer normalization on the output state of the bidirectional GRU layer;
and S63, repeating the steps S61 to S62, and stacking two layers of bidirectional GRU coding networks.
8. The method for predicting traffic speed based on the neural network of the time chart according to claim 1 or 3, wherein the step S7 specifically comprises:
s71, calculating attention values between the traffic state at the current moment and all observation states of the encoding end;
and S72, splicing the attention result, the input layer node characteristics and the output state of the bidirectional GRU layer, outputting the prediction value matrixes of all nodes at the current moment through a full-connection layer, and summarizing to obtain a prediction result sequence with the length of T'.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110417357.XA CN113159414B (en) | 2021-04-19 | 2021-04-19 | Traffic speed prediction method based on timing diagram neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110417357.XA CN113159414B (en) | 2021-04-19 | 2021-04-19 | Traffic speed prediction method based on timing diagram neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113159414A true CN113159414A (en) | 2021-07-23 |
| CN113159414B CN113159414B (en) | 2022-05-24 |
Family
ID=76868447
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110417357.XA Active CN113159414B (en) | 2021-04-19 | 2021-04-19 | Traffic speed prediction method based on timing diagram neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113159414B (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113627591A (en) * | 2021-07-30 | 2021-11-09 | 马上消费金融股份有限公司 | Dynamic graph data processing method and device, electronic equipment and storage medium |
| CN113794695A (en) * | 2021-08-26 | 2021-12-14 | 浙江工业大学 | GRU-based network abnormal flow detection and identification method |
| CN114511767A (en) * | 2022-02-11 | 2022-05-17 | 电子科技大学 | Quick state prediction method for timing diagram data |
| CN115906928A (en) * | 2022-11-25 | 2023-04-04 | 中国矿业大学 | Transformer UUV three-dimensional autonomous collision avoidance planning method based on double-channel self-attention |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109919358A (en) * | 2019-01-31 | 2019-06-21 | 中国科学院软件研究所 | A kind of real-time site traffic prediction technique based on neural network space-time attention mechanism |
| CN110648527A (en) * | 2019-08-20 | 2020-01-03 | 浙江工业大学 | Traffic speed prediction method based on deep learning model |
| CN112241814A (en) * | 2020-10-20 | 2021-01-19 | 河南大学 | Traffic prediction method based on reinforced space-time diagram neural network |
| CN112667920A (en) * | 2020-12-29 | 2021-04-16 | 复旦大学 | Text perception-based social influence prediction method, device and equipment |
-
2021
- 2021-04-19 CN CN202110417357.XA patent/CN113159414B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109919358A (en) * | 2019-01-31 | 2019-06-21 | 中国科学院软件研究所 | A kind of real-time site traffic prediction technique based on neural network space-time attention mechanism |
| CN110648527A (en) * | 2019-08-20 | 2020-01-03 | 浙江工业大学 | Traffic speed prediction method based on deep learning model |
| CN112241814A (en) * | 2020-10-20 | 2021-01-19 | 河南大学 | Traffic prediction method based on reinforced space-time diagram neural network |
| CN112667920A (en) * | 2020-12-29 | 2021-04-16 | 复旦大学 | Text perception-based social influence prediction method, device and equipment |
Non-Patent Citations (1)
| Title |
|---|
| 薛声军: ""基于时空图卷积网络的城市路网交通流预测研究"", 《中国优秀硕士学位论文全文数据库（电子期刊）》 * |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113627591A (en) * | 2021-07-30 | 2021-11-09 | 马上消费金融股份有限公司 | Dynamic graph data processing method and device, electronic equipment and storage medium |
| CN113794695A (en) * | 2021-08-26 | 2021-12-14 | 浙江工业大学 | GRU-based network abnormal flow detection and identification method |
| CN114511767A (en) * | 2022-02-11 | 2022-05-17 | 电子科技大学 | Quick state prediction method for timing diagram data |
| CN114511767B (en) * | 2022-02-11 | 2023-12-01 | 电子科技大学 | Rapid state prediction method for time sequence diagram data |
| CN115906928A (en) * | 2022-11-25 | 2023-04-04 | 中国矿业大学 | Transformer UUV three-dimensional autonomous collision avoidance planning method based on double-channel self-attention |
| CN115906928B (en) * | 2022-11-25 | 2023-05-16 | 中国矿业大学 | Two-channel self-attention-based transform UUV three-dimensional autonomous collision avoidance planning method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113159414B (en) | 2022-05-24 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113159414B (en) | Traffic speed prediction method based on timing diagram neural network | |
| CN110648527B (en) | Traffic speed prediction method based on deep learning model | |
| CN112863180B (en) | Traffic speed prediction method, device, electronic equipment and computer readable medium | |
| CN114299723B (en) | Traffic flow prediction method | |
| CN115240425A (en) | Traffic prediction method based on multi-scale space-time fusion graph network | |
| CN115273464A (en) | Traffic flow prediction method based on improved space-time Transformer | |
| CN111260919B (en) | Traffic flow prediction method | |
| CN115578851A (en) | Traffic prediction method based on MGCN | |
| Rathore et al. | Exploiting real-time big data to empower smart transportation using big graphs | |
| CN109284682A (en) | A kind of gesture identification method and system based on STT-LSTM network | |
| CN114021809A (en) | Unmanned driving is with picture level convolution neural network model of traffic flow prediction | |
| CN113762338A (en) | Traffic flow prediction method, equipment and medium based on multi-graph attention mechanism | |
| CN115936069A (en) | Traffic flow prediction method based on space-time attention network | |
| CN114493034B (en) | Regional flow prediction-oriented space-time global semantic representation learning method | |
| CN113627676A (en) | Traffic prediction method and system based on multi-attention causal relationship | |
| CN113821985B (en) | Traffic state prediction method and device and electronic equipment | |
| CN115762147A (en) | Traffic flow prediction method based on adaptive graph attention neural network | |
| CN115168327A (en) | Large-scale data space-time prediction method based on multilayer tree long-short term memory network | |
| CN115985086A (en) | Traffic data completion method, system, terminal and storage medium | |
| CN116129646B (en) | Traffic prediction method of graph convolution neural network based on feature intersection | |
| CN114360242A (en) | Urban short-term traffic flow prediction method and system based on hybrid convolution LSTM | |
| Chen et al. | Traffic flow prediction based on cooperative vehicle infrastructure for cloud control platform | |
| Li et al. | MTGCN: Multi-graph Fusion Based Temporal-Spatial Convolution for Traffic Flow Forecasting | |
| CN118247953A (en) | Traffic flow prediction method and device by combining rainfall and space-time diagram convolution model | |
| CN117218868B (en) | Traffic flow prediction method based on geometric scattering map network |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |